Vision based features in moisture content measurement during raisin production

Abstract

Apparent characteristics of fruits are different in various moisture contents. Some feature changes can be quantified in different stages of drying process. Study on these changes is useful to determine the qualitative properties of dried product. This article represents an approach to investigate the effects of dehydrating on some apparent characteristics of raisin comprising shape, color and texture which are interpretable by machine vision systems. The correlation between the time and the image based characteristics were correlated in raisin drying. Samples were dried under 50°C in an oven dryer. Variations in five shape, nine color and five texture features corresponding to the moisture contents of the kernels were measured at different time intervals during the drying process. All morphological features decreased smoothly with drying time. Several ANN models were employed to predict the moisture content. Overall performance of the network was verified using the excluded test data. Results showed that the machine vision system integrated with the neural networks is a proper tool to predict the moisture content with acceptable values of coefficient of determination (R² = 99.84%), Root Mean Squared Error (RMSE = 0.00078) and Mean Absolute (MAPE = 0.0312%).